Newey-West HAC Standard Errors
Also known as: HAC standard errors, Heteroskedasticity and Autocorrelation Consistent covariance, Bartlett kernel HAC estimator, HAC düzeltmeli standart hatalar
Newey-West HAC standard errors, introduced by Whitney Newey and Kenneth West in 1987, provide a covariance matrix estimator for OLS regression that remains valid under both heteroskedasticity and serial autocorrelation of unknown form. They are the standard tool for correcting inference in time-series and panel regression when residuals are not i.i.d., requiring no specification of the error structure beyond choosing a bandwidth parameter.
Read the full method
Sign in with a free account to read this section.
Method map
The neighbourhood of related methods — select a node to explore.
When to use it
Use Newey-West HAC standard errors whenever OLS residuals from a time-series or panel regression display serial autocorrelation, heteroskedasticity, or both, and the error structure is unknown. The estimator is consistent as sample size grows provided the bandwidth L grows at rate T^(1/4). It is less appropriate when cross-sectional dependence dominates (prefer cluster-robust errors) or when parametric FGLS is available. It does not correct bias in point estimates — only inference.
Strengths & limitations
- Valid under both heteroskedasticity and autocorrelation of unknown form without specifying an error model.
- Positive semi-definiteness is guaranteed by the Bartlett kernel, ensuring well-defined standard errors.
- Widely implemented in standard econometric software (R, Stata, Python statsmodels, EViews) with minimal user effort.
- Asymptotically consistent under mild regularity conditions as bandwidth grows appropriately with sample size.
- Finite-sample performance can be poor when the bandwidth L is chosen too small or too large relative to the degree of autocorrelation.
- Addresses only time-series dependence; cross-sectional clustering requires separate cluster-robust estimators.
- Does not correct point estimates — if errors are serially correlated, OLS is inefficient and FGLS may be preferable.
- Automatic bandwidth selection rules (e.g., Andrews 1991) may still undercover in highly persistent processes.
Frequently asked
How do I choose the bandwidth L?
The original paper recommends L proportional to T^(1/4). Data-driven methods, such as the Andrews (1991) plug-in rule, estimate the optimal bandwidth from the data by fitting an AR(1) or other parametric model to the residuals. Many software packages implement automatic bandwidth selection by default.
Does Newey-West fix autocorrelation in my model?
No. HAC standard errors adjust hypothesis tests and confidence intervals for autocorrelation but leave OLS point estimates unchanged. If residuals are autocorrelated, OLS remains consistent under exogeneity but is inefficient; FGLS or GLS would produce more precise estimates. HAC is a diagnostic aid, not a structural cure.
When should I use cluster-robust rather than HAC standard errors?
Use cluster-robust errors when the dominant dependence is across observations within groups (e.g., individuals within firms or regions) rather than across time. In panel data with both cross-sectional and temporal dependence, two-way cluster-robust estimators or Driscoll-Kraay standard errors may be more appropriate than HAC alone.
Sources
- Newey, W. K., & West, K. D. (1987). A simple, positive semi-definite, heteroskedasticity and autocorrelation consistent covariance matrix. Econometrica, 55(3), 703–708. DOI: 10.2307/1913610 ↗
How to cite this page
ScholarGate. (2026, June 2). Newey-West HAC Standard Errors. ScholarGate. https://scholargate.app/en/econometrics/newey-west-hac
Which method?
Set this method beside its closest kin and read them side by side — the library lays the books on the table; the choice is yours.
- OLS RegressionEconometrics↔ compare